Following #65 , we also wanted to update the survival results to use the new data shuffling approach. Instead, we actually ended up dropping the shuffled models for survival (they tend to have issues with convergence) and deciding just to plot raw concordance index values. As a baseline/"negative control", we're instead using a model that only uses clinical covariates (age, sample mutation burden, sample cancer type).
Overall, our models tend to (slightly) outperform the baseline (grey dotted line), but there's not much difference between -omics types:
Code changes:
Added survival results QC script: 06_predict_survival/nbconverted/survival_results_qc.py (this is new)
Moved 06_predict_survival/nbconverted/plot_survival_pc_comparison.py to 06_predict_survival/nbconverted/plot_survival_results.py and made a few changes to generate paper figure versions with clinical-only baselines
Renamed 06_predict_survival/nbconverted/plot_survival_heatmaps.py (this has already been reviewed, you don't need to look at it again)
Update scripts in 06_predict_survival to run covariate-only baseline
PR overview:
Following #65 , we also wanted to update the survival results to use the new data shuffling approach. Instead, we actually ended up dropping the shuffled models for survival (they tend to have issues with convergence) and deciding just to plot raw concordance index values. As a baseline/"negative control", we're instead using a model that only uses clinical covariates (age, sample mutation burden, sample cancer type).
Overall, our models tend to (slightly) outperform the baseline (grey dotted line), but there's not much difference between -omics types:
Code changes:
06_predict_survival/nbconverted/survival_results_qc.py(this is new)06_predict_survival/nbconverted/plot_survival_pc_comparison.pyto06_predict_survival/nbconverted/plot_survival_results.pyand made a few changes to generate paper figure versions with clinical-only baselines06_predict_survival/nbconverted/plot_survival_heatmaps.py(this has already been reviewed, you don't need to look at it again)06_predict_survivalto run covariate-only baseline